OpenAI ended 2025 with roughly 7,850 employees and plans to reach 8,000 by the end of 2026, up from 770 in late 2023. Its average employee equity package, $1.5 million in stock, is the largest ever recorded at a private tech startup. This post covers OpenAI’s headcount, pay, department mix, office locations, and hiring data for 2026. OpenAI Employee Statistics 2026 – TL;DR OpenAI is the largest independent AI lab by headcount, roughly twice the size of Anthropic. The numbers below answe...

Odd Scenarios about Research Claims I blogged about OpenAI's achievement of having AI solve a math problem  here . My post had a few comments about authorship of such results. Lance had a post about co-authorship and AI  here There are times when an author on a paper prefers to not be listed as a co-author. There are times when an author wants to give credit in odd places. We give some scenarios. 1) Professors Alice and Bob  prove a theorem. They have their grad student Carol write up the res...

How to create an organic gradient animation using a WebGL shader. How to create an organic gradient animation using a WebGL shader.

As companies adopt AI tools, a lot of time is spent on thinking about AI policies from a security, compliance, or even cost-focused angle. But many leaders are neglecting to address how their teams should work with AI in the context of the team as a whole. This creates a lot of unresolved tension, and it’s time for leaders to step up and set some guidelines not just for how to use AI in an “approved” sense, but how to use it respectfully. When I say respectfully, I am not talking about the...

House building using the “Oraaflex” modular construction system, via Wikipedia . Welcome to the reading list, a weekly roundup of news and links related to buildings, infrastructure and industrial technology. This week we look at a California chemical leak, weapons-grade plutonium for nuclear reactor startups, a startup that will clean your house to get robot training data, Blue Origin’s rocket explosion, and more. Roughly 2/3rds of the reading list is paywalled, so for full access become...

Drive-by AI changes break the shared model a team builds around its code, and the ICs end up cleaning up the mess. Why pushing to mainline should come with the pager. Drive-by AI changes break the shared model a team builds around its code, and the ICs end up cleaning up the mess. Why pushing to mainline should come with the pager.

Summary: People have tried extending rock-paper-scissors to more than 3 options, with some success. If you allow more pairings to be ties, this uncovers a rich garden of different game dynamics and strategies. As far as a battle of wits go, it’s hard to find a more balanced game than rock-paper-scissors. Its simplicity means that all strategy is scraped away, and all that’s left is to cold read the other person’s soul. Come, stare into my eyes, and I’ll see the weapon you’ll ...

Hello, It has been some time, and we're happy to share some news with you today. 2.1 plan We have been working on the major 2.1 update for Factorio and Space age for the last 8 months, and it has been shaping up quite well. Scope and expectations Generally, we are happy with the game design of Factorio and Space age. The progression is good, things are mostly well balanced (one or two exceptions), and there isn't anything we feel is majorly missing. That is to say, we d...

In recent weeks, I have been talking with a lot of people about personal websites. In so many of my discussions, I mention that one of the reasons I love coming back to my personal website is the community around the indie web: people all over the world sharing what interests them: slices of life, hopes and dreams, tutorials on how to do something, and more. Websites aren’t islands. Websites are houses in a town. That brings me to JUnited . I first participated in JUnited , a challenge that ...

Last week I posted about why the accept attribute on file inputs is bad UX. The accept attribute lets you specify which file types an input will accept. For example, if users need to upload a receipt, you can do this: That means users can only select those file types. All other files are disabled: This sounds good because it stops users from picking an invalid type before they submit - which would then cause an error. But as I said in the post, this is bad UX because: T...

Unary codes are a form of universal variable-length code (UVLC) that are sometimes used on their own but more commonly used as a building block for more general families of UVLCs like Golomb, Rice or Gamma / Exp-Golomb codes. There are multiple conventions in use, the one I’ll use in the following has codewords terminated with a “1” bit and is 0-based, i.e. the codebook goes 0 -> 1 1 -> 01 2 -> 001 3 -> 0001 4 -> 00001 and so forth. That is, some value i ...

Clara and I finally tied the knot after more than a decade. It was a short ceremony at the Pyrmont registry with close family, but it was lovely. The celebrant joked that it was “her first Sanrio wedding” on account of all of Clara’s Pompompurins , many of which her brother and fiancé brought for her. It was adorable. We met at our university anime club . We hung out, and gradually it turned into something more. Our first “date” involved us sitting at a bubble tea shop and talki...

I was thinking of doing something special for the 100th post, but I could not decide what to do, so instead I'll do just some random bits and pieces. I've been living on the Internet since the early 00s and during that time I had many different blogs and blog-like sites. Some disappeared of natural causes, some can still be found. A photo from one of my blog posts from 2009. This is my laptop of that time, a HP Compaq with which I have very fond memories. It's placed on the floor of the ro...

In 1980, Intel introduced the 8087 floating-point chip, a co-processor that made floating-point operations up to 100 times faster. This chip was highly influential, and today most processors use the floating-point standard introduced by the 8087. The 8087 uses complicated algorithms to accurately compute functions such as square roots, tangents, and exponentials. These algorithms are implemented inside the chip in low-level code called microcode. I'm part of a group, the Opcode Collective, tha...

This was posted at about 01:30am so it's probably full or errors but whatever I never back out of writing a post all in one go. If I've spoken to you in the past week, I probably mentioned playing around with Plan 9 . After getting an old PC on ebay for 30€ (~40€ with shipping) I thought "if I was ever going to install this, now is the time". Shout out to Matt for the hint of what to look for to get real cheap used PCs on ebay; which we talked about a bit on an IndieWeb pop-up about B...

Where is your intelligence located? In your brain? It is a simplistic answer. A better model is that your intelligence is embodied. Consider a cook working at an expensive restaurant. He has all his favorite knives and cooking instructions, placed exactly where he wants them. His kitchen is part of his intelligence, of his skills. The same cook working in your kitchen can probably cook better than you do, but he can’t reproduce the same meals he would prepare in his favorite kitchen. We ...

Where are all the AI-generated projects? This is a common question from AI skeptics: if LLMs are so good at writing code, where is the tsunami of new AI-generated apps, services and games? I personally don’t find this to be much of a paradox. Writing code is only one of the bottlenecks involved in actually shipping a new product, after all. It’s also impossible to talk about the paid work I’ve done with AI (you’ll simply have to take my word that it’s increased my productivity). ...

It has been an unfortunate turn in the software industry, one of many as of late, that gambling is once again one of its primary engines. With the rise of almost nationwide online sports betting, not to mention prediction markets, making odds on real-world events and extracting the money of suckers is no longer limited to island nations. It is a great American pursuit, or at least, that's what modern television sports coverage leads you to believe. There has always been an uncomfortable relati...

A single server handles jobs that arrive randomly and take a random amount of time to process. If both inter-arrival times and service times follow exponential distributions, this is called an M/M/1 queue , and is the simplest model in queueing theory. Managers often treat utilization linearly: “90% busy is only a little worse than 80% busy.” The M/M/1 formula shows this intuition is badly wrong. The mean number of jobs in the system (waiting and being served) is: $$L = \frac{\rho}{1 - ...

Here we go again. Afternoon walk this time around. It’s almost 2pm, and I’m standing in the same parking spot where I got picked up last week. No breakfast in me, but I did have lunch before heading out. Compared to last week’s hike, this one’s gonna be way easier. We have a bit more than 20kms to walk, with roughly 650 meters of ascent and 1300 of descent. Gonna be fun. Before we begin, I’ll have to apologise for the terrible photos I took, especially of the churches. Been a weird w...

When I’ve read other series about Kubernetes and reach the secrets section my eyes glaze over. I can’t help myself; I want to read about the fun stuff. Secrets are necessary to be sure, but it’s a little boring… But if I want to do proper GitOps I need to manage secrets (and to document the process). The sooner I set it up the better. In and outside cluster Kubernetes has different solutions for secrets management. Of particular note is Sealed Secrets which creates files that a...

My nephew just graduated high school, and wants a laptop. When he decides what computer to buy, price (or more precisely, value ) is the most important attribute. Apple's MacBook Neo upended the 'value laptop' equation—Apple's not supposed to be both the cheapest option and the best value... but it seems like that's squarely where the Neo landed for the good-but-cheap laptop category. My nephew is also my godson, and to kick off his computing journey, I thought I'd let him choose f...

Anthropic shipped Claude Opus 4.8 today. My favourite thing about it is this note in the release announcement: Users will find Opus 4.8 to be a modest but tangible improvement on its predecessor. There’s still more to be done: we’re working on developing and releasing models that provide many of the same capabilities as Opus at a lower cost. It's so refreshing to see an AI lab honestly describe a release as a minor incremental improvement over the previous model! Honesty seems to...